Backtesting forex: specifics and pitfalls

Forex trades 24 hours a day on weekdays, with no single opening bell: the same pair behaves nothing like it does at 3 a.m. versus the London open. A backtest that lumps all hours together hides this reality.

Another quirk: you trade pairs (EUR/USD, GBP/JPY…), often correlated. Two “different” positions can actually carry the same disguised risk if the pairs move together.

On forex, the main cost is the spread, and it isn't constant: it widens at the open, on news, and in the thin Asian session. Backtesting with a tiny fixed spread almost always overstates performance.

For small-target strategies (scalping), the spread alone can flip expectancy from positive to negative. Model a realistic spread — ideally variable — plus any swap (overnight fee) if you hold positions overnight.

The London and New York sessions concentrate most of the volume and directional moves; Asia is often quieter and range-bound. A breakout strategy works better in an active session, a range strategy in a quiet one.

Testing with and without a session filter, over a large sample, reveals whether your edge depends on a specific window — useful to know, but beware over-optimizing a time window on history.

Replay bar by bar (or tick by tick) on real forex data, including spread and costs, never seeing the future. Log each trade's session to analyze where your strategy truly performs.

On GetBacktest, replay runs on real data with the future hidden server-side, and the robustness verdict (Monte-Carlo, walk-forward, deflated Sharpe) tells you whether your forex edge survives out-of-sample.